Simulation Based Expert System to Predict the Tensile Behaviour of Tailor Welded Blanks

Abhishek T. Dhumal, R. Ganesh Narayanan

Department of Mechanical Engineering, Indian Institute of Technology Guwahati, Guwahati 781 039, India. E-mail: ganu@iitg.ernet.in

G. Saravana Kumar
Department of Engineering Design, Indian Institute of Technology Madras, Chennai 600 036, India. E-mail: gsaravana@iitm.ac.in
Abstract: The forming behaviour of Tailor Welded Blanks (TWB) is influenced by thickness ratio, strength ratio, and weld conditions in a synergistic fashion. In most of the cases, these parameters deteriorate the forming behaviour of TWB. It is necessary to predict suitable TWB conditions for achieving better stamped product made of welded blanks. This is quite difficult and resource intensive, requiring lot of simulations or experiments to be performed under varied base material and weld conditions. Automotive sheet part designers will be greatly benefited if an expert system is available that can deliver forming behaviour of TWB for varied weld and blank conditions. This work primarily aims at developing an expert system based on Artificial Neural Network (ANN) model to predict the tensile behaviour of welded blanks made of steel grade and aluminium alloy base materials. The important tensile characteristics of TWB are predicted within wide range of varied blank and weld conditions. The tensile test (sample with a notch) is simulated in finite element code ABAQUS 6.7 for generating the required output data for ANN modelling. Predicted results namely limit strains, failure location, minimum thickness, and strain path from ANN model are compared and validated with simulation results for two different intermediate TWB conditions. It is observed that the results obtained from ANN are encouraging with acceptable prediction errors. This expert system module is part of the global expert system framework that is proposed [1, 2] for designing TWB conditions that will deliver better TWB products. Keywords: Tailor welded blanks, tensile test simulation, finite element method, expert system, neural network.

1. INTRODUCTION Tailor Welded Blanks (TWB) are blanks of similar or dissimilar thicknesses, materials, coatings etc. welded in a single plane before forming. This welded blank is then formed like un-welded blanks to manufacture automotive components, with appropriate tooling and forming conditions. Applications of TWB include car door inner panel, deck lids, bumper, side frame rails etc. in automotive sector [3,4]. The forming

behaviour of TWBs is critically influenced by thickness and material combinations of the blanks welded; weld conditions like weld orientation, weld location, and weld properties in a synergistic manner and hence it is difficult to design the TWB conditions that can deliver a good stamped product with more or less similar formability characteristics as that of un-welded blank. In this context, few research groups have aimed at predicting the formability of welded blanks by using

Volume 13 Issue 1 2011 IJAMS

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

different necking criteria based on simulation work. For example, Anand et al. [5] investigated the limit strains of welded blanks made of IF steel of different thicknesses (0.75-1.5 mm). The modified Keeler criterion was used to predict the plane-strain limit strain (FLD0) in which average thickness and average strain hardening exponent of the blanks constituting TWB was considered. The predicted FLD0 is comparable with that of from experiments. Davies et al. [6] investigated the limit strains of aluminium alloy TWB (1:2 mm thickness), where in the Failure Limit Curve (FLC) predicted by Marciniak-Kuczynski (M-K) analysis were compared with the experimental results. Here the geometrical heterogeneity, i.e., the initial imperfection level, involved in the welded blank was modeled by using the strain-hardening exponent determined from miniature tensile testing together with the Hosford yield criterion, involving parameters a = 8 and R = 1 to determine a level of imperfection that exactly fits an FLD to each experimentally evaluated failure strains. The FLCs thus predicted are found to have good agreement with the experimental FLCs, except in the bi-axial stretching region. An interesting work was done by Lee et al. [7] in predicting the forming limit and load-stroke behaviour of Friction Stir Welded (FSW) blanks. In this investigation, wide variety of automotive sheet materials viz., 6111-T4, 5083-H18, 5083-O aluminium alloys, dual-phase steel (DP590) and AZ31 magnesium alloy sheets were experimentally tested and their forming limit were predicted using M-K model. The predictions are in good agreement with the experimental FLCs. Ganesh and Narasimhan [8] predicted the forming limit strains of laser welded blanks by using thickness gradient based necking criterion incorporated into a FE simulation code PAMSTAMP 2G. It is found that the predictions are good in drawing region of FLD, with deviation in stretching region. From the above discussion, it is clear that one has to follow a limit strain theory in
160

conjunction with numerical or analytical methods to predict the forming limit strains of welded blanks for different base material and weld conditions. Similarly tensile behaviour of welded blanks was also predicted for varied base metal and weld conditions. For example, Ganesh and Narasimhan [9, 10] studied the influence of varied weld conditions on the tensile and forming behaviour of welded blanks by numerical simulations. Here the main aim is to identify the sub domain of weld conditions with in which weld zone assumption is required instead of weld line assumption. Also the relative effect of TWB parameters on the forming behaviour was studied. Similar work has been performed by Raymond et al. [11] also. It is clear from the above discussion that, in general, either numerical simulation or analytical models are used to predict the tensile behaviour of welded blanks. The work performed by Chen et al. [12] to understand the cold roll forming process and Yang at al. [13] to study the effect of rolling parameters on the rolling process are other examples where numerical simulations are used to understand the forming behaviour. It is known that the presence of thickness, strength heterogeneities and weld region deteriorates the formability of welded blanks in most of the cases. Designing TWB for a typical application will be successful only by knowing the appropriate thickness, strength combinations, weld line location and profile, number of welds, weld orientation and weld zone properties. Predicting these TWB parameters in advance will be helpful in determining the formability of TWB part in comparison to that of un-welded base materials. In order to fulfill this requirement, one has to perform lot of simulation and experimental trials separately for each of the cases which is time consuming and resource intensive. Stamping designers will be greatly benefited if an expert system is available for TWBs that can deliver its forming behaviour

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

for varied weld and blank conditions. The significance of expert systems in manufacturing process control is appreciated from the following literature; in sheet forming [14], casting defects prediction [15], and in other areas like structural design [16], accident prevention [17] etc. Artificial Neural Network (ANN) is found to show better prediction of any response variable that is influenced by large number of input parameters rather than conventional way of doing experiments. Application of ANN technique in developing an expert system for TWB formability prediction will definitely be helpful in understanding and designing the TWB conditions that can deliver a better-stamped product. The main objective of the present research scheme is to develop an expert system for welded blanks that can predict their tensile, deep drawing, forming behaviour under varied base material and weld conditions using different formability tests, material models, and formability criteria. The expert system will be developed in conjunction with neural network. This expert system/ANN model can be associated to a fully automated deep drawing machine, a mechatronic system, in which it acts as brain controlling the whole process. For example, the expert system can predict the strain path for the given input properties, by which the deformation force can be varied suitably to minimize thinning beyond a limit and the occurrence of limit strains leading to failure. In the present work, a simulation based expert system is developed to predict the tensile behaviour of TWBs. The data required for expert system development is generated by simulation trials only. Global TWB tensile behaviour like limit strains, failure location, minimum thickness, and strain path are predicted for a wide range of thickness and strength combinations, weld properties and orientation. The accuracy of expert system thus developed is validated with simulation results for chosen intermediate levels. This
161

expert system will be integrated with the global expert system described in Veerababu et al. [1, 2] for predicting the formability of TWBs. The numerical simulation work based on finite elements is also used to understand the general forming process characteristics. 2. METHODOLOGY The flow chart describing the methodology followed in this work is shown in Figure 1. The first part of methodology involves simulation design and deals with the design of series of input property combinations and their values to generate required data for expert system development. In order to conduct the exercise with optimum simulations, Design Of Experiments (DOE) using the Taguchis statistical design [18] is followed. Simulation models for predicting the tensile behaviour of TWBs are constructed as per the DOE parameter tables. The second part of the methodology is the ANN modelling and validation. The post processed results of FE simulations are used to train the ANN. Finally the ANN model/expert system is validated with simulation results for chosen intermediate levels. It should be noted that the present work will lead to addition of expert system module is the expert system proposed and described in Veerababu et al. [1,2] work. The methodology is discussed in the following subsections. 2.1 Selection of process parameters and their levels for tensile behaviour prediction 2.1.1 Base material properties and TWB parameters Initially for conducting simulation trials the material and process parameters that affect the TWB tensile behaviour are identified from available literature. TWBs made of steel and aluminium alloy sheets are considered as base materials. Table 1 shows the mechanical and forming properties of base metal and weld.

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

The plastic strain ratios of weld zone are assumed to be one in all the rolling directions as it is assumed isotropic. In order to generate the required data for expert system with optimum simulations, the Taguchis statistical design is followed. In view of cost saving and time restriction higher order interactions are neglected. The six factors considered at three levels are shown in Table 2 for steel and aluminium alloy TWBs. The TWB parameters considered for the analysis are (1) thickness ratio, (2) yield strength ratio, (3) weld orientation, (4) weld n value, (5) weld yield strength and (6) weld width. The schematic representation of these parameters is depicted in Figure 2 and Figure 3b. Each parameter has three levels (1, 2 and 3). The levels of parameters are chosen in such a way that the range covers practically all the combinations in typical experiments and industrial parts [1, 19, 20]. The average thickness of thinner and thicker sheets is assumed as weld zone thickness in simulation trials. 2.1.2 Selection of orthogonal array Taguchi statistical design is a powerful tool to screen the significant experiments from many by conducting relatively lesser number of experiments. In this work, L27 orthogonal array with linear graph indicating the allocation of individual factors in orthogonal array is followed. Here L27 orthogonal array corresponds to three levels with six factors. Material properties

Identifying the parameters affecting the formability of welded blanks and their levels

Weld conditions viz., weld Weld zone Base width, orientation, location properties material properties DOE for training and testing simulation trails, selecting full factorial and orthogonal array Modeling simulation of tensile tests Creating CAD models Meshing and creating material data base Define and apply material properties Apply boundary conditions Post processing and analyzing results Test data Training data Artificial Neural Network (ANN) modeling Define neural network architecture Fix error goals and other training parameters Train the network for minimum error Acceptable No Yes Validate ANN results with simulation results

Figure 1: Overall methodology of simulation and developing expert system

Aluminium alloy sheet Base metal Weld zone 77 77 2700 2700 0.3 0.3 0.7 1 0.6 1 0.8 1 0.172 See Table 2

Steel sheet [8] Base metal Weld zone Youngs modulus (E), GPa 210 210 3 Density (), kg/m 7860 7860 Poissons ratio () 0.3 0.3 r0 1.21 1 r45 1.08 1 r90 1.68 1 Strain hardening exponent (n) 0.27 See Table 2

Table 1: Material properties of steel grade and aluminium alloy base materials
162

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

Parameters/levels Thickness ratio (T1/T2), mm/mm Strength ratio (YS1/YS2), MPa/MPa Weld orientation() Weld n value (nw) Weld yield strength, (YSw), MPa Weld width (W), mm

Aluminium alloy base metal [21] 1 2 3 1 2 3 0.5 0.75 1 0.5 0.75 1 (.75/1.5) (1.125/1.5) (1.5/1.5) (0.75/1.5) (1.125/1.5) (1.5/1.5) 0.5 0.75 1 0.5 0.75 1 (150/300) (225/300) (300/300) (190/380) (285/380) 380/380) 0 45 90 0 45 90 0.1 0.15 0.2 0.1 0.13 0.15 Steel grade base metal 125 2 250 5 500 10 150 2 300 5 400 10

Table 2: TWB parameters for steel grade and aluminium alloy TWB and their levels
Thickness ratio, TR = T1 / T2 Yield strength ratio, SR = YS1 / YS2 Weld strain hardening exponent, nw Weld yield strength, YSw Weld width, W T2, YS2 T1, YS1

Thinner blank

Weld zone (W, YSw, nw) Thicker blank

Figure 2: Schematic representation of controllable factors

However, this design fundamentally does not account for the interaction among the processing parameters. In view of cost and time saving, these interactions have been neglected. If necessary, the missing interactions can be analyzed by further running the required experiments. Since L27 orthogonal array is followed, 27 simulations are performed to generate data for ANN modelling. The tensile behaviour, viz., limit strain, failure location, minimum thickness, and strain path are predicted for each test simulation. The limit strains are predicted by thickness gradient based necking criterion. 2.2 Modelling simulation of tensile test for welded blanks CAD models of tensile specimen are
163

generated as per geometry shown in Figure 3a [22] in Pro-E and imported into ABAQUS 6.7 for preprocessing, performing simulations and post processing. The meshing is done with quadrilateral shell elements having five through-thickness integration points. In this automatic structured meshing technique is used. The meshed blank thus obtained is divided into three different regions viz., weld region (without HAZ), base material 1 and base material 2 to construct meshed models of TWB for varied weld orientations. A constant mesh size of 1 mm is kept in the weld region and base metal [8]. The material properties are assigned to weld zone and base metals according to the different parameter levels in the orthogonal array. Displacement boundary conditions (Figure 3b) are applied to the tensile sample such that one end of the specimen is fixed and the other end is given finite displacement with a velocity of 0.5mm/min. In order that the TWB fails during simulation, a geometrical notch of 10 mm width is provided. This notch geometry is decided based on trial simulations such that the entire deformation is concentrated only in that region and finally necking occurs, without much deformation happening in the shoulder region. For this, varied notch widths 14 mm, 10 mm, and 8 mm were simulated and compared with each other.

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

12 55

10

16

a)

90 Notch Shoulder

45

b)

Figure 3: a) Schematic of representation of tensile sample modeled in ABAQUS 6.7; b) TWB sample with different weld orientations and boundary conditions ( Displacement; Not to scale; Dimensions in mm) TWB sample with different weld orientations and boundary)
Finally the notch of 10 mm width is selected, wherein the effect of different TWB factors is not suppressed because of the notch effect and lesser deformation is observed in the shoulder region during simulations. This notched tensile sample is used to simulate TWB tensile test for data generation. Swift law ( = K (0 + p)n, K strength co-efficient, n strain hardening exponent, 0 pre-strain value of 0.003) is used as strain hardening law describing the stress-strain relationship of weld and base material. It should be noted that an un-notched specimen, as well as, Hollomon law was used by Veerababu et al. [1] to describe the constitutive behaviour of TWB. Hills 1948 isotropic hardening yield criterion [23] is used as the plasticity model for both steel and aluminium alloy base materials. After simulations, five output parameters are predicted for steel and aluminium TWB. These parameters are described below. Limit strain (major and minor strain): As per thickness gradient criterion, necking occurs when the thickness ratio between thinner and thicker element reaches 0.92 [24]. The thinner element has already failed and hence can not be referred for limit strain prediction. This means that the strain in the
164

thinner element is above actual limit strain value. So the thicker element which is closer to thinner element is referred for the prediction work. This procedure is followed for all the 27 simulations trials. The limit strains are found to be in negative minor strain region of FLD (Figure 4), because of the presence of notch and tensile, plane-strain strain paths. Failure location: It is the distance from the fixed end to the thicker element in the progression where necking has occurred or criterion is satisfied. Minimum thickness: It is the minimum thickness of the element of specimen in the progression where necking has occurred. Strain path: It is the plot between major and minor strain from the starting progression to the progression where necking has occurred. This is quantified by the slope of the strain path curve (Figure 4). 2.3 ANN modelling of tensile behaviour of TWBs In the present work, an ANN is trained to

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

learn arbitrary nonlinear relationships between input and used to obtain the deformation behaviour of TWB for any given input property combinations. The various ANN parameters like number of hidden layers, neurons, and transfer functions are optimized based on many trials to predict the outputs within the normalized error limit of 10-4.
0.4 necking point or limit strain 0.35 0.3
Major strain

intermediate TWB parameter levels (between level one and level two and level two and level three) are utilized for testing. The intermediate input levels chosen for validation purpose are given in Table 5 for tensile test simulation and prediction by ANN.
Output Number of hidden layers 2 2 2 2 1 Number of neurons in each layer 12, 6 12, 6 8, 4 8, 4 6

0.25 0.2 0.15 0.1 Strain path 0.05 0 -0.3 -0.2 -0.1 0 Minor strain

Forming limit curve

Forming starting

Major strain Minor strain Failure location Minimum thickness Strain path slope
0.3

0.1

0.2

Figure 4: Schematic representation of strain path with forming limit curve

Various network architectures with one and two hidden layers with varying number of neurons in each layer are examined and compared with each other. The final ANN architecture contains 6 input neurons (corresponding to 6 factors) and 1 output neuron (corresponding to 1 of the 5 outputs to be predicted) with tan sigmoid and pure linear as transfer functions. Five ANNs each for steel and aluminium were considered to predict the five outputs and their architecture was optimized by varying the number of hidden layers and neurons in each layer. The final optimized architecture for different output parameters are listed in Table 3 for steel based TWB and Table 4 for aluminium alloy based TWB. A feed forward back propagation algorithm is selected to train the network in Matlab programming environment. Here the scaled conjugate gradient algorithm is used to minimize the error. From the available simulation data sets, 27 data sets are used to train and two
165

Table 3: ANN architectures used for tensile simulation of TWB with steel base material
Output Number of hidden layers 2 2 2 2 1 Number of neurons in each layer 12, 6 12, 6 6, 3 6, 3 6

Major strain Minor strain Failure location Minimum thickness Strain path slope

Table 4: ANN architectures used for tensile simulation of TWB with aluminium alloy base material
3. RESULTS AND DISCUSSION In this work, six different TWB parameters listed in Table 2 are varied at few levels by following L27 orthogonal array and the important tensile behaviours namely limit strain, strain path, minimum thickness and

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

failure location are predicted using simulation. The data generated are used for ANN training and validation. 3.1 General behaviour observation of tensile

Figure 5 a, b shows the 27 limit strain data for steel and aluminium TWBs respectively. It is clear that in both the TWBs, though tensile test is simulated, most of the limit strains are close to plane-strain strain path (i.e., major strain axis) and not in tensile strain path. This is mainly because of the notch present in the tensile sample that is used for simulation for failure occurrence. In the case of aluminium TWB, more limit strain values are clustered near to plane-strain condition as compared to steel TWB. Another important observation is that the strain path slope varies from 18.77 to 0.379 in the case of steel TWB and as high as 200.33 to 1.938 in the case of aluminium TWB. The lower slope values correspond to limit strain values that are away from planestrain condition. In the case of steel TWB, the limit strains values vary from (0.677, -0.2074) to (0.089, 0.0076). In this, the maximum and minimum limit strain values correspond to experiments 20 and 9 in L27 orthogonal array of experiments respectively. The experiment 20 corresponds to thickness ratio of 1 and has

base material failure as compared to experiment 9 that is characterized by weld failure and thickness ratio of 0.5. In a similar fashion, for aluminium TWB, the maximum limit strain is characterized by thickness ratio and strength ratio equal to 1 and longitudinal weld orientation. The failure location values for different experiments are shown in Figure 6 for steel and aluminium TWBs. The failure is expected to occur within the notch region, either in the base material or in the weld region depending on the weld width. It is clear from Figure 6 that in almost all cases failure has occurred within the span, except in few cases wherein failure is seen just outside the notch region. The failure location is found to show significant effect on the minimum thickness achieved during TWB forming. Figure 7 shows the variation of minimum thickness achieved for different experiments. In the case of steel TWB, minimum thickness of 0.3492 mm occurs in experiment 15, for which failure location is at 66.22 mm from the gripping end. This location is close to edge of the notch depicted as AA in Figure 6 (see inset). Similarly in aluminium TWB, a minimum thickness of 0.1729 mm occurs in experiment 5, for which the failure location is at 65.93 mm, which is also close to the notch edge AA in Figure 6.

Aluminium alloy base metal Parameters/levels 1 2 1 2 Thickness ratio (T1/T2), mm/mm 0.7 (1.05/1.5) 0.9 (1.35/1.5) 0.7 (1.05/1.5) 0.9 (1.35/1.5) Strength ratio (YS1/YS2), MPa/MPa 0.7 (210/300) 0.9 (270/300) 0.6 (180/300) 0.8 (240/300) Weld orientation() 35 55 35 55 Weld n value (nw) 0.165 0.18 0.12 0.135 Weld yield strength, (YSW), MPa 225 400 175 325 Weld width (W), mm 4 8 4 8 Steel grade base metal

Table 5: Input levels for validating the ANN prediction of TWB tensile behaviour

166

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

1.2 1

1.2 1

Major strain

0.6 0.4 0.2 0 -0.5 -0.4 -0.3 -0.2 Minor strain -0.1 0

0.6 0.4 0.2 0 -0.5 -0.4 -0.3 -0.2 Minor strain -0.1 0

a) Steel TWB

b) Aluminium TWB

Figure 5: Limit strain values for different TWB condition

3.2 Prediction of tensile behaviour by ANN The limit strain (major and minor strain), failure location, minimum thickness, strain path are predicted using the trained ANN and validated with FE simulation results for two intermediate input levels (Table 5), both for steel and aluminium alloy TWB. These input values are incorporated into FE simulations for validation purpose. Tables 6 and 7 show the comparison between ANN and simulation results for TWB made of steel and aluminium alloy base materials respectively for all the five output factors. The ANN prediction results are comparable to that of from simulations for same TWB conditions. The error statistics pertaining to the ANN modelling of tensile test is given in Tables 8 and 9. A percentage error of less than 6 % is considered as acceptable in the present domain [8]. In the case of steel TWB, strain path slope (7.6 %) and failure location (6.52 %) in the case of aluminium alloy TWB have more percentage error than this limit. All other parameters show better prediction level with acceptable error range. The proposed ANN based expert system is in early stages of development. The expert system is envisaged to be expanded with industrial applications also. For example, a sheet forming engineer who wants to develop
167

expert system for some industrial TWB sheet part can just make it as part of existing system framework in the same line of thought, without introducing new rules and conditions. The relations between TWB inputs and outputs are non-linear in nature and hence it is complex to explicitly state rules for making expert system. But these complex relationships can be captured by ANN. Since this expert system is based on ANN, it is a continuous learning system as the problem solved by the system can also become a part of training examples. Though the expert system can not reason out the decisions/results unlike rule based systems, one can interpret the results by comparing the outputs of two different input conditions quantitatively with minimum knowledge in TWB forming behaviour. The ANN learning and arriving at optimum architecture initially takes several iterations and is also problem specific. The expert system/ANN model developed in this work is applicable within the range of input and base material properties specified in Tables 1 and 2. Though this is true, the range specified in Table 2 is large enough to include usable TWB conditions. It is worth to study the applicability of the present expert system outside this range and for many new sheet materials including high strength steels.

Major strain

0.8

0.8

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

80

Failure location outside notch

70
Failure location (mm)

60 50 40 30 20 10 0 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 Experiment number

Failure location within notch

12 55
Steel TWB Aluminium TWB

A
10

16

Figure 6: Failure location for different simulation trials for steel and aluminium TWB

Figure 7: Variation of minimum thickness achieved for different simulation trial

Output Test Data 1 FE simulation 0.485 0.087 73.9 0.517 3.23 ANN prediction 0.502 0.091 69.4 0.492 3.49 Test Data 2 FE simulation 0.600 0.068 69.5 0.790 5.22 ANN prediction 0.620 0.071 65.2 0.752 4.82

Major strain Minor strain Failure location, mm Minimum thickness, mm Strain path slope

Table 6: Validation of prediction by ANN for steel TWB

Output Test Data 1
168

Test Data 2

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

Major strain Minor strain Failure location, mm Minimum thickness, mm Strain path slope

FE simulation 0.887 0.054 69.1 0.105 2.24

ANN prediction 0.935 0.055 73.9 0.109 2.33

FE simulation 0.356 0.150 73.5 0.279 2.71

ANN prediction 0.375 0.146 68.7 0.291 2.61

Table 7: Validation of prediction by ANN for aluminium TWB

Output Training % error Major strain 0.0043 Minor strain 0.0078 Failure location 0.0053 Minimum thickness 0.0024 Strain path slope 1.45 Testing SD in error % error 0.081 3.3 0.2 4.19 0.029 6.12 0.0015 4.8 0.43 7.6

SD in error 0.1863 0.3349 4.5685 1.4741 2.38

Table 8: Error values for training and testing of ANN (Steel base material)
Output Training % error Major strain 0.0075 Minor strain 0.067 Failure location 0.071 Minimum thickness 0.0028 Strain path slope 0.052 Testing % error 5.23 2.79 6.52 4.28 3.91

SD in error 0.23 0.92 0.76 0.031 0.417

SD in error 3.511 0.87 1.82 2.64 0.36

Table 9: Error values for training and testing of ANN (Aluminium alloy base material)
4. CONCLUDING REMARKS The following are the conclusions made from the present work: Both the steel and aluminium TWBs show strain path close to plane strain condition, though tensile test was simulated. This is mainly due to the presence of notch in the tensile sample. TWBs with more thickness, strength heterogeneities and weld failure show less limit strains as compared to others out of all 27 different welded blanks properties. The failure location is found to occur within the notch in most of the TWB cases. An expert system is developed to predict the tensile behaviour of tailor welded blanks. This expert system has ability to successfully predict the tensile behaviour like limit strains, failure location, minimum thickness, and strain path.

REFERENCES [1] Veera Babu K., Ganesh Narayanan R., & Saravana Kumar G., An expert system based on artificial neural network for predicting the tensile behaviour of tailor welded blanks,
169

ABHISHEK T. DHUMAL, R. GANESH NARAYANAN, AND G. SARAVANA KUMAR

Expert Systems with Applications, 2009, 36, 1068310695. [2] Veera Babu K., Ganesh Narayanan R., & Saravana Kumar G., An expert system for predicting the deep drawing behaviour of tailor welded blanks, Expert Systems with Applications, 2010, 37, 7802-7812. [3] Kusuda H., Takasago T., & Natsumi F., Formability of tailored blanks, Journal of Materials Processing Technology, 1997, 71, 134140. [4] Pallet R. J., & Lark R. J., The use of tailored blanks in the manfacture of construction components, Journal of Materials Processing Technology, 2001, 117, 249-254. [5] Anand D., Boudreau G., Andreychuk P., Chen D. L., & Bhole S. D., Forming behaviour of tailor (laser) welded blanks of automotive steel sheet, The Canadian Journal of Metallurgy and Materials Science, 2006, 45(2), 187-197. [6] Davies R. W., Smith M. T., Oliver H. E., Khaleel M. A., & Pitman S. G., Weld metal ductility in aluminium tailor welded blanks, Metallurgical and Materials Transactions A, 2000, 31(11), 2755-2763. [7] Lee W., Chung K-H., Kim D., Kim J., Kim C., Okamoto K., Wagoner R. H., & Chung K., Experimental and numerical study on formability of friction stir welded TWB sheets based on hemispherical dome stretch tests, International Journal of Plasticity, 2009, 25(9), 1626-1654. [8] Ganesh Narayanan R., & Narasimhan K., Predicting the forming limit strains of tailor welded blanks, Journal of Strain Analysis for Engineering Design, 2008, 43(7), 551-563. [9] Ganesh Narayanan R., & Narasimhan K., Weld region representation during the simulation of TWB forming behaviour, International Journal of Forming Processes, 2006, 9(4), 491-518. [10] Ganesh Narayanan, R., & Narasimhan, K., Relative effect of material and geometric parameters on the forming behaviour of tailor

welded blanks (TWB), International Journal of Forming Processes, 2007, 10(2), 145-178. [11] Raymond S. C., Wild P. M., & Bayley C. J., On modelling of the weld line in finite element analyses of tailor-welded blank forming operations, Journal of Materials Processing Technology, 2004, 147(1), 28-37. [12] Chen X., Chang Q., Wang J., & Chen C., Simulation of cold roll forming using elasticplastic finite element method, International Journal of Modelling, Identification and Control, 2009, 7(1), 97 102. [13] Yang L., Hu J., Ning L., & Liu Y., Research on influence of rolling parameters on the rolling process based on numerical simulation, International Journal of Modelling, Identification and Control, 2009, 7(1), 25 32. [14] Palani R., Wagoner R. H., & Narasimhan K., Intelligent design environment: A knowledge based simulations approach for sheet metal forming, Journal of Materials Processing Technology, 1994, 45, 703-708. [15] Karunakar B. D., & Datta G. L., Prediction of defects in castings using back propagation neural networks, International Journal of Modelling, Identification and Control, 2008, 3(2), 140 147. [16] Mukherjee A., & Deshpande J. M., Application of artificial neural networks in structural design expert systems, Computers & Structures, 1995, 54(3), 367-375. [17] Chiou Y-C., An artificial neural networkbased expert system for the appraisal of twocar crash accidents, Accident Analysis and Prevention, 2006, 38, 777785. [18] Taguchi G., 1990, Introduction to Quality Engineering, Asian Productivity Organization, Tokya. [19] Stasik M. C., & Wagoner R. H., Forming of tailor welded aluminium blanks, International Journal of Forming Processes, 1998, 1(1), 9-33. [20] Miles M. P., Decker, B. J., & Nelson, T.W., Formability and strength of friction-stirwelded aluminium sheets, Metallurgical and

170

SIMULATION BASED EXPERT SYSTEM TO PREDICT THE TENSILE BEHAVIOUR OF TAILOR WELDED BLANKS

Material transactions A, 2004, 35(11), 34613468. [21] Liu S., & Chao Y. J., Determination of global mechanical response of friction stir welded plates using local constitutive properties, Modelling and simulation in materials science and engineering, 2005, 13(1), 115. [22] Holmberg S., Enquist B., & Thilderkvist P., Evaluation of sheet metal formability by tensile tests, Journal of Materials Processing Technology, 2004, 145, 72-83. [23] Banabic, D. et al., 2000, Formability of Metallic Materials, Springer, Berlin. [24] Sujit Kumar, Date P. P., & Narasimhan K., A new criterion to predict necking failure under biaxial stretching, Journal of Materials Processing Technology, 1994, 45, 583-588. BIOGRAPHIES Abhishek T. Dhumal received his M. Tech. (Machine Design) degree from Department of Mechanical Engineering, Indian Institute of Technology Guwahati, India in 2009. Dr. R. Ganesh Narayanan is presently an Assistant Professor at Department of Mechanical Engineering, Indian Institute of Technology Guwahati, India. His area of research includes sheet metal forming and cold forging, tailor welded blanks, formability and forming limit, and computational applications in sheet forming. He completed Doctoral degree from Department of Metallurgical Engineering and Materials Science, Indian Institute of Technology Bombay, India in 2007 and Master degree from Manufacturing Engineering division, College of Engineering, Anna University, Chennai, India. Dr. G. Saravana Kumar is an assistant professor of Department of Engineering Design at Indian Institute of Technology, Madras, India. He has a masters degree in Welding Engineering from Department of
171

Metallurgical and Materials Engineering, National Institute of Technology, Tiruchirappalli, India and received his Ph.D. degree from Department of Mechanical Engineering, Indian Institute of Technology, Kanpur, India in 2004. His research interests include computer aided design and manufacturing, computational geometry, reverse engineering, rapid prototyping, soft computing techniques and expert systems in engineering.